Problem API

`UQPyL.problem`

The problem module defines the shared modeling protocol used by sampling, analysis, optimization, inference, calibration, and surrogate modeling.

Import

python

from UQPyL.problem import Problem, ModelProblem, Eval

Public Objects

Object	Role
`Problem`	Define a static objective and optional constraint problem.
`ModelProblem`	Define a simulation problem with optional observations and masks.
`ModelEvalContext`	Runtime context passed to `ModelProblem` objective, constraint, or evaluation callables.
`Eval`	Standard return object from `evaluate()`.
`Space`	Define a bounded input space with continuous, integer, or discrete variables.
`SpaceBase`	Base input-space interface.
`ProblemBase`	Base problem interface used by built-in benchmark problems.
`ProblemABC`	Alias of `ProblemBase`.
`singleFunc`	Decorator for adapting one-sample objective functions to batched input.
`singleEval`	Decorator for adapting one-sample `Eval` functions to batched input.

`Problem`

Use Problem when objectives and constraints are computed directly from input variables.

python

Problem(
    nInput=None,
    nObj=None,
    ub=None,
    lb=None,
    objFunc=None,
    conFunc=None,
    evaluate=None,
    conWgt=None,
    nCon=0,
    varType=None,
    varSet=None,
    optType="min",
    xLabels=None,
    name=None,
    space=None,
    objLabels=None,
    conLabels=None,
)

Parameter	Meaning
`nInput`	Number of input variables. Required unless `space` is provided.
`nObj`	Number of objectives. Required.
`ub`	Upper bounds. Scalar, list, or NumPy array.
`lb`	Lower bounds. Scalar, list, or NumPy array.
`objFunc`	Objective callable. Receives batched `X`, returns shape `(n_samples, n_obj)`.
`conFunc`	Constraint callable. Receives batched `X`, returns shape `(n_samples, n_con)`.
`evaluate`	Combined evaluation callable. Must return `Eval`.
`conWgt`	Optional constraint weights.
`nCon`	Number of constraints. Defaults to `0`.
`varType`	Variable type list. `0` continuous, `1` integer, `2` discrete.
`varSet`	Discrete value mapping for variables with `varType=2`.
`optType`	Optimization direction. `"min"`, `"max"`, or one value per objective.
`xLabels`	Optional input labels.
`name`	Optional problem name.
`space`	Optional `SpaceBase` object. When provided, it defines the input space.
`objLabels`	Optional objective labels.
`conLabels`	Optional constraint labels.

Problem accepts exactly one callable configuration:

Configuration	Use when
`objFunc`	The problem has objectives only.
`objFunc` + `conFunc`	The problem has objectives and constraints.
`evaluate`	Objectives and constraints should be computed together.

Do not combine evaluate with objFunc or conFunc.

Methods

Method	Returns	Meaning
`evaluate(X, target=None)`	`Eval`	Evaluate objectives and constraints.
`objFunc(X)`	`np.ndarray`	Evaluate objectives only.
`conFunc(X)`	`np.ndarray` or `None`	Evaluate constraints only.
`validate(X)`	`np.ndarray`	Convert input to 2D and check dimension.
`unit_to_space(X, IFlag=True, DFlag=True)`	`np.ndarray`	Map unit-space values to problem bounds.
`apply_var_type(X, IFlag=True, DFlag=True)`	`np.ndarray`	Apply integer and discrete variable transforms.
`cast_int_vars(X)`	`np.ndarray`	Round integer variables.
`map_discrete_vars(X)`	`np.ndarray`	Map discrete variables to values in `varSet`.
`getOptimum()`	implementation-specific	Return a known optimum when implemented.

target can be:

Value	Meaning
`None`	Return all available outputs.
`"objs"`	Return objectives only.
`"cons"`	Return constraints only.

Example:

python

import numpy as np

from UQPyL.problem import Problem


def objFunc(X):
    X = np.atleast_2d(X)
    return np.sum(X**2, axis=1, keepdims=True)


problem = Problem(
    nInput=2,
    nObj=1,
    ub=1.0,
    lb=-1.0,
    objFunc=objFunc,
)

res = problem.evaluate([[0.2, 0.3]])
print(res.objs)

`ModelProblem`

Use ModelProblem when the primary callable is a simulation model.

python

ModelProblem(
    nInput=None,
    nObj=1,
    ub=None,
    lb=None,
    simFunc=None,
    objFunc=None,
    conFunc=None,
    evaluate=None,
    obs=None,
    mask=None,
    conWgt=None,
    nCon=0,
    varType=None,
    varSet=None,
    optType="min",
    xLabels=None,
    name=None,
    space=None,
    objLabels=None,
    conLabels=None,
    simLabels=None,
)

Parameter	Meaning
`simFunc`	Required simulation callable. Receives batched `X`; returns numeric NumPy array whose first dimension is `n_samples`.
`objFunc`	Optional objective callable. Receives `(X, context)`.
`conFunc`	Optional constraint callable. Receives `(X, context)`.
`evaluate`	Optional combined callable. Receives `(X, context)` and returns `Eval`.
`obs`	Optional 2D observation array with shape `(n_time, n_series)`.
`mask`	Optional boolean mask with the same shape as `obs`.
`simLabels`	Optional labels for simulation series.

Other parameters are the same as Problem.

Methods

Method	Returns	Meaning
`evaluate(X, target=None)`	`Eval`	Evaluate simulation, objectives, and constraints.
`simFunc(X)`	`np.ndarray`	Run the simulation callable and validate output.
`buildContext(X)`	`ModelEvalContext`	Run simulation and build the context object.
`objFunc(X, context=None)`	`np.ndarray`	Evaluate objectives. Builds context when omitted.
`conFunc(X, context=None)`	`np.ndarray` or `None`	Evaluate constraints. Builds context when omitted.
`flattenSim(sim)`	`np.ndarray`	Flatten simulation output to `(n_samples, n_obs)`.
`flattenObs()`	`np.ndarray`	Flatten observations to `(n_obs,)`.
`flattenMask()`	`np.ndarray`	Flatten mask to `(n_obs,)`.

ModelProblem.evaluate() supports one extra target:

Value	Meaning
`"sim"`	Return simulation output only in `Eval.sim`.

Example:

python

import numpy as np

from UQPyL.problem import ModelProblem


obs = np.array([[1.0], [2.0]])


def simFunc(X):
    X = np.atleast_2d(X)
    sim = np.zeros((X.shape[0], 2, 1))
    sim[:, 0, 0] = X[:, 0]
    sim[:, 1, 0] = X[:, 1]
    return sim


def objFunc(X, context):
    err = context.sim - context.obs
    return np.mean(err**2, axis=(1, 2)).reshape(-1, 1)


problem = ModelProblem(
    nInput=2,
    nObj=1,
    ub=3.0,
    lb=0.0,
    simFunc=simFunc,
    objFunc=objFunc,
    obs=obs,
)

res = problem.evaluate([[1.0, 2.2]])
print(res.objs)
print(res.sim)

`Eval`

Eval is the standard return object from evaluate().

python

Eval(objs=None, cons=None, sim=None)

Field	Type	Meaning
`objs`	`np.ndarray` or `None`	Objective values.
`cons`	`np.ndarray` or `None`	Constraint values. Feasible constraints satisfy `cons <= 0`.
`sim`	`np.ndarray` or `None`	Simulation output for `ModelProblem`.

Property	Meaning
`hasObjs`	`True` when `objs` is not `None`.
`hasCons`	`True` when `cons` is not `None`.
`hasSim`	`True` when `sim` is not `None`.

`Space`

Space defines input dimension, bounds, labels, and variable type transformations.

python

Space(
    nInput,
    ub,
    lb,
    varType=None,
    varSet=None,
    xLabels=None,
)

Parameter	Meaning
`nInput`	Number of input variables.
`ub`	Upper bounds. Scalar, list, or NumPy array.
`lb`	Lower bounds. Scalar, list, or NumPy array.
`varType`	Optional variable type list.
`varSet`	Required mapping for discrete variables.
`xLabels`	Optional input labels.

Variable types:

Value	Meaning
`0`	Continuous
`1`	Integer
`2`	Discrete

Methods:

Method	Returns	Meaning
`validate(X)`	`np.ndarray`	Convert input to 2D and check dimension.
`transform(X)`	`np.ndarray`	Validate input and apply variable type transforms.
`unit_to_space(X, IFlag=True, DFlag=True)`	`np.ndarray`	Map unit-space values to bounds.
`apply_var_type(X, IFlag=True, DFlag=True)`	`np.ndarray`	Apply integer and discrete transforms.
`cast_int_vars(X)`	`np.ndarray`	Round integer variables.
`map_discrete_vars(X)`	`np.ndarray`	Map discrete variables to `varSet` values.

Attributes:

Attribute	Meaning
`nInput`	Number of input variables.
`ub`	Upper bounds as shape `(1, n_input)`.
`lb`	Lower bounds as shape `(1, n_input)`.
`varType`	Variable type array.
`idxF`	Continuous variable indices.
`idxI`	Integer variable indices.
`idxD`	Discrete variable indices.
`varSet`	Discrete variable value mapping.
`xLabels`	Input labels.
`encoding`	`"real"` for continuous spaces, `"mix"` for mixed spaces.

Decorators

`singleFunc`

Use singleFunc when an objective function is easier to write for one sample at a time.

python

from UQPyL.problem import Problem, singleFunc


@singleFunc
def objFunc(x):
    return x[0] ** 2 + x[1] ** 2


problem = Problem(nInput=2, nObj=1, ub=1.0, lb=-1.0, objFunc=objFunc)

The wrapped function receives one 1D sample and the wrapper returns a 2D objective array.

`singleEval`

Use singleEval when a one-sample function returns Eval.

python

import numpy as np

from UQPyL.problem import Eval, Problem, singleEval


@singleEval
def evaluate(x):
    return Eval(
        objs=np.array([x[0] ** 2 + x[1] ** 2]),
        cons=np.array([x[0] + x[1] - 1.0]),
    )


problem = Problem(
    nInput=2,
    nObj=1,
    nCon=1,
    ub=1.0,
    lb=0.0,
    evaluate=evaluate,
)

Benchmark Problems

Built-in benchmark problems can be imported from UQPyL.problem.

Single-Objective Problems

Class
`Sphere`
`Schwefel_2_22`
`Schwefel_1_22`
`Schwefel_2_21`
`Rosenbrock`
`Step`
`Quartic`
`Schwefel_2_26`
`Rastrigin`
`Ackley`
`Griewank`
`Trid`
`Bent_Cigar`
`Discus`
`Weierstrass`
`RosenbrockWithCon`

Multi-Objective Problems

Class
`ZDT1`
`ZDT2`
`ZDT3`
`ZDT4`
`ZDT6`
`DTLZ1`
`DTLZ2`
`DTLZ3`
`DTLZ4`
`DTLZ5`
`DTLZ6`
`DTLZ7`

Example:

python

from UQPyL.problem import Sphere, ZDT1


sphere = Sphere(nInput=10)
zdt1 = ZDT1(nInput=30)

print(sphere.evaluate([[0.0] * 10]).objs)
print(zdt1.evaluate([[0.5] * 30]).objs)

Problem API

Problem API ​

UQPyL.problem ​

Import ​

Public Objects ​

Problem ​

Methods ​

ModelProblem ​

Methods ​

Eval ​

Space ​

Decorators ​

singleFunc ​

singleEval ​

Benchmark Problems ​

Single-Objective Problems ​

Multi-Objective Problems ​

Problem API

`UQPyL.problem`

Import

Public Objects

`Problem`

Methods

`ModelProblem`

Methods

`Eval`

`Space`

Decorators

`singleFunc`

`singleEval`

Benchmark Problems

Single-Objective Problems

Multi-Objective Problems